Use octet_length for Hyper, Umbra, CedarDB

[vogelsgesang]

Switch Hyper, Umbra, and CedarDB to octet_length so Q27 and Q28 measure bytes, matching the intended ClickBench semantics while avoiding UTF-8 character scans when byte lengths are available from string offsets.

The queries for ClickHouse use length(URL) / length(Referer). In ClickHouse, length is byte-oriented (https://clickhouse.com/docs/sql-reference/functions/string-functions#length), in contrast to lengthUTF8. Given ClickHouse itself uses a byte-length, the intent of ClickBench was probably to measure bytes, not UTF8. Hyper / Umbra / CedarDB were accidentally using UTF8 lengths, since their length function aligns with Postgres and nobody noticed the difference.

Also see #981, #985

[vogelsgesang]

CC @toschmidt re the changes to Umbra
CC @lukasvogel re the changes to CedarDB

Please let know if you disagree with those changes. Happy to back them out of this PR, if you don't want them

[github-actions]

Results for cedardb are ready for: c6a.4xlarge.
Results for cedardb-parquet are ready for: c6a.4xlarge.
Results for hyper are ready for: c6a.4xlarge.
Results for hyper-parquet-partitioned are ready for: c6a.4xlarge.
Results for hyper-parquet-single are ready for: c6a.4xlarge.
Results for umbra are ready for: c6a.4xlarge.
The result files are committed as c355792.

Logs:

• umbra on c6a.4xlarge: benchmark log, cloud-init log
• cedardb-parquet on c6a.4xlarge: benchmark log, cloud-init log
• cedardb on c6a.4xlarge: benchmark log, cloud-init log
• hyper-parquet-partitioned on c6a.4xlarge: benchmark log, cloud-init log
• hyper-parquet-single on c6a.4xlarge: benchmark log, cloud-init log
• hyper on c6a.4xlarge: benchmark log, cloud-init log

[github-actions]

Results for cedardb are ready for: c6a.metal.
The result files are committed as a6a1000.
The run of cedardb on t3a.small did not produce results.
The run of hyper on c6a.large did not produce results.
The run of hyper on c8g.4xlarge did not produce results.
The run of hyper on c8g.metal-48xl did not produce results.
The run of hyper on t3a.small did not produce results.
The run of hyper-parquet-partitioned on c8g.4xlarge did not produce results.
The run of hyper-parquet-partitioned on c8g.metal-48xl did not produce results.
The run of hyper-parquet-single on c8g.4xlarge did not produce results.
The run of hyper-parquet-single on c8g.metal-48xl did not produce results.

Logs:

• cedardb on t3a.small: benchmark log, cloud-init log
• hyper on c8g.4xlarge: benchmark log, cloud-init log
• hyper-parquet-partitioned on c8g.4xlarge: benchmark log, cloud-init log
• hyper-parquet-single on c8g.4xlarge: benchmark log, cloud-init log
• hyper on c8g.metal-48xl: benchmark log, cloud-init log
• hyper-parquet-partitioned on c8g.metal-48xl: benchmark log, cloud-init log
• hyper-parquet-single on c8g.metal-48xl: benchmark log, cloud-init log
• hyper on c6a.large: benchmark log, cloud-init log
• hyper on t3a.small: benchmark log, cloud-init log
• cedardb on c6a.metal: benchmark log, cloud-init log

[github-actions]

Results for cedardb are ready for: c6a.4xlarge, c6a.xlarge, c7a.metal-48xl, c8g.4xlarge, c8g.metal-48xl.
Results for cedardb-parquet are ready for: c6a.2xlarge, c6a.4xlarge, c6a.large, c6a.metal, c6a.xlarge, c7a.metal-48xl, c8g.4xlarge, c8g.metal-48xl, t3a.small.
Results for hyper are ready for: c6a.2xlarge, c6a.4xlarge, c6a.metal, c6a.xlarge, c7a.metal-48xl.
Results for hyper-parquet-partitioned are ready for: c6a.2xlarge, c6a.4xlarge, c6a.large, c6a.metal, c6a.xlarge, c7a.metal-48xl.
Results for hyper-parquet-single are ready for: c6a.2xlarge, c6a.4xlarge, c6a.large, c6a.metal, c6a.xlarge, c7a.metal-48xl.
Results for umbra are ready for: c6a.2xlarge, c6a.4xlarge, c6a.large, c6a.metal, c6a.xlarge, c7a.metal-48xl, c8g.4xlarge, c8g.metal-48xl, t3a.small.
The result files are committed as 46596a6.

Logs:

• cedardb on c7a.metal-48xl: benchmark log, cloud-init log
• cedardb on c8g.metal-48xl: benchmark log, cloud-init log
• umbra on c6a.metal: benchmark log, cloud-init log
• umbra on c6a.4xlarge: benchmark log, cloud-init log
• umbra on c7a.metal-48xl: benchmark log, cloud-init log
• hyper-parquet-partitioned on c6a.metal: benchmark log, cloud-init log
• cedardb-parquet on c6a.metal: benchmark log, cloud-init log
• hyper-parquet-partitioned on c7a.metal-48xl: benchmark log, cloud-init log
• umbra on c8g.metal-48xl: benchmark log, cloud-init log
• umbra on c8g.4xlarge: benchmark log, cloud-init log
• cedardb-parquet on c7a.metal-48xl: benchmark log, cloud-init log
• hyper-parquet-single on c6a.metal: benchmark log, cloud-init log
• cedardb-parquet on c8g.metal-48xl: benchmark log, cloud-init log
• hyper-parquet-single on c7a.metal-48xl: benchmark log, cloud-init log
• umbra on c6a.2xlarge: benchmark log, cloud-init log
• cedardb-parquet on c8g.4xlarge: benchmark log, cloud-init log
• cedardb-parquet on c6a.4xlarge: benchmark log, cloud-init log
• cedardb on c6a.4xlarge: benchmark log, cloud-init log
• hyper-parquet-single on c6a.4xlarge: benchmark log, cloud-init log
• hyper-parquet-partitioned on c6a.2xlarge: benchmark log, cloud-init log
• cedardb-parquet on c6a.xlarge: benchmark log, cloud-init log
• hyper-parquet-single on c6a.2xlarge: benchmark log, cloud-init log
• umbra on c6a.xlarge: benchmark log, cloud-init log
• umbra on c6a.large: benchmark log, cloud-init log
• cedardb on c8g.4xlarge: benchmark log, cloud-init log
• cedardb-parquet on c6a.large: benchmark log, cloud-init log
• hyper on c7a.metal-48xl: benchmark log, cloud-init log
• hyper on c6a.metal: benchmark log, cloud-init log
• hyper-parquet-partitioned on c6a.4xlarge: benchmark log, cloud-init log
• cedardb-parquet on c6a.2xlarge: benchmark log, cloud-init log
• hyper on c6a.2xlarge: benchmark log, cloud-init log
• hyper-parquet-single on c6a.xlarge: benchmark log, cloud-init log
• umbra on t3a.small: benchmark log, cloud-init log
• hyper-parquet-partitioned on c6a.xlarge: benchmark log, cloud-init log
• cedardb-parquet on t3a.small: benchmark log, cloud-init log
• hyper on c6a.4xlarge: benchmark log, cloud-init log
• hyper-parquet-single on c6a.large: benchmark log, cloud-init log
• hyper-parquet-partitioned on c6a.large: benchmark log, cloud-init log
• hyper on c6a.xlarge: benchmark log, cloud-init log
• cedardb on c6a.xlarge: benchmark log, cloud-init log

[github-actions]

Results for hyper-parquet-partitioned are ready for: t3a.small.
Results for hyper-parquet-single are ready for: t3a.small.
The result files are committed as 026bf5a.

Logs:

• hyper-parquet-partitioned on t3a.small: benchmark log, cloud-init log
• hyper-parquet-single on t3a.small: benchmark log, cloud-init log

[github-actions]

Results for cedardb are ready for: c6a.2xlarge.
The result files are committed as ab0e1d5.

Logs:

• cedardb on c6a.2xlarge: benchmark log, cloud-init log

[github-actions]

The run of cedardb on c6a.large did not produce results.

Logs:

• cedardb on c6a.large: benchmark log, cloud-init log

[vogelsgesang]

just for reference (and so I can find it again later):

c6a.4xlarge, median of the two warm runs:

Database	Q27	Q28	Aggregate effect
Hyper	458 → 146 ms (3.14×)	7.109 → 6.912 s (2.8%)	~2.7%
Umbra	290 → 115 ms (2.53×)	1.526 → 1.397 s (8.5%)	~2.3%
CedarDB	275 → 108 ms (2.54×)	3.198 → 3.097 s (3.2%)	~2.2%